Blanchet , J. and Kang , Y. (2020). Semi-supervised Learning Based on Distributionally Robust Optimization. In Data Analysis and Applications 3 (eds A. Makrides, A. Karagrigoriou and C.H. Skiadas). https://doi.org/10.1002/9781119721871.ch1
Abstract
This chapter proposes a novel method for semi‐supervised learning (SSL) based on data‐driven distributionally robust optimization (DRO) using optimal transport metrics. The proposed method enhances generalization error by using the non‐labeled data to restrict the support of the worstcase distribution in DRO formulation. The chapter describes the implementation of DRO formulation by proposing a stochastic gradient descent algorithm, and demonstrates that semi‐supervised DRO method is able to improve the generalization error over natural supervised procedures and state‐of‐the‐art SSL estimators. It includes a discussion on the large sample behavior of the optimal uncertainty region in the DRO formulation. The discussion exposes important aspects such as the role of dimension reduction in SSL.
Authors
Jose Blanchet, Yang Kang
Publication date
2020/4/15
Journal
Data Analysis and Applications 3: Computational, Classification, Financial, Statistical and Stochastic Methods
Volume
5
Pages
1-33
Publisher
John Wiley & Sons, Inc.
